<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[mcpprobe Research]]></title><description><![CDATA[Research and practical guides on Model Context Protocol (MCP) server security: what AI-agent connectors expose, what the numbers say, and how to check your own ]]></description><link>https://mcpprobe.hashnode.dev</link><image><url>https://cdn.hashnode.com/uploads/logos/6af3ad1517fcddd5d4a97a9f/bcbbd90b-82d9-4ff4-9ce3-fb4f595f1182.png</url><title>mcpprobe Research</title><link>https://mcpprobe.hashnode.dev</link></image><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 15:06:00 GMT</lastBuildDate><atom:link href="https://mcpprobe.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Five checks to run on your MCP server before someone else does]]></title><description><![CDATA[55.7% of public MCP servers answer with no login. Five quick curl checks to see what your own MCP server exposes before someone else finds out.]]></description><link>https://mcpprobe.hashnode.dev/five-checks-to-run-on-your-mcp-server-before-someone-else-does</link><guid isPermaLink="true">https://mcpprobe.hashnode.dev/five-checks-to-run-on-your-mcp-server-before-someone-else-does</guid><category><![CDATA[mcp]]></category><category><![CDATA[Security]]></category><category><![CDATA[AI]]></category><category><![CDATA[oauth]]></category><category><![CDATA[Devops]]></category><dc:creator><![CDATA[Atabek]]></dc:creator><pubDate>Thu, 12 Nov 2026 23:24:09 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af3ad1517fcddd5d4a97a9f/f50115d4-b4ba-4196-bc9f-d8d60a15188f.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>We recently surveyed the public servers listed in the official MCP registry. We didn't call any tools; we only read what each server tells an anonymous client on first contact. More than half of the servers we could assess (55.7%) answered with no login at all. That isn't always wrong. But it's worth knowing which side of the line your own server is on.</p>
<p>Here are five checks you can run against a server <strong>you own</strong> in about ten minutes, using nothing but curl.</p>
<p>Set your endpoint first:</p>
<pre><code class="language-bash">MCP=https://your-server.example/mcp
</code></pre>
<h2>1. Does it answer without a login?</h2>
<pre><code class="language-bash">curl -s -X POST "$MCP" \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"self-check","version":"0"}}}' \
  -i | head -20
</code></pre>
<p>A <code>401</code> is what you want if the server touches anything private. A <code>200</code> with a <code>serverInfo</code> block means anyone on the internet can talk to it. That's fine for a public docs-search server, and not fine for anything else.</p>
<h2>2. If it requires login, can a client find out how?</h2>
<p>The MCP authorization spec asks protected servers to publish OAuth protected resource metadata (RFC 9728). Check both places a client looks:</p>
<pre><code class="language-bash">curl -si "$MCP" -X POST -d '{}' | grep -i www-authenticate
curl -s https://your-server.example/.well-known/oauth-protected-resource | head -c 500
</code></pre>
<p>You want a <code>WWW-Authenticate</code> header that carries a <code>resource_metadata</code> URL, or a JSON document at the well-known path listing your <code>authorization_servers</code>. In our survey, <strong>15.1% of servers that required login published neither</strong>. A standards-following client can't sign in to them, and the usual "fix" is to switch auth off, which is worse.</p>
<h2>3. What does an anonymous client see in your tool list?</h2>
<p>If check 1 returned 200, list the tools the same way (first send <code>notifications/initialized</code> and reuse the session header if your server issues one). Then read the result as an attacker would:</p>
<ul>
<li>Is there anything that runs commands, reads or writes files, or touches a database? In our survey, <strong>102 open servers</strong> offered exactly that to anyone.</li>
<li>Is every tool marked with <code>readOnlyHint</code> or <code>destructiveHint</code>? Only <strong>19.6%</strong> of open servers labelled all of their tools. Without those hints, a client can't tell when to ask the user first.</li>
</ul>
<h2>4. Are your tool descriptions written for people?</h2>
<p>The model reads tool descriptions as instructions. Search yours for:</p>
<ul>
<li>phrases addressed to the model ("do not tell the user", "always call this first", "ignore previous");</li>
<li>invisible characters, such as zero-width spaces or bidi controls;</li>
<li>long hidden blocks after the visible text.</li>
</ul>
<pre><code class="language-bash"># paste your tools/list JSON into tools.json, then:
grep -nP '[\x{200B}-\x{200F}\x{202A}-\x{202E}\x{2060}-\x{2064}\x{FEFF}]' tools.json
</code></pre>
<p>One caution from our own work: zero-width joiners are legitimate inside Persian, Arabic and Indic text and in emoji sequences. A match isn't automatically an attack.</p>
<h2>5. Would you notice if a description changed?</h2>
<p>Most clients approve a server once. Nothing in the protocol stops the server serving different tool definitions later. Keep a hash of what you shipped:</p>
<pre><code class="language-bash">jq -S '.result.tools' tools.json | sha256sum
</code></pre>
<p>Store it in CI. If it changes without a matching pull request, find out why.</p>
<h2>The short version</h2>
<ol>
<li>Require login for anything that reads private data or changes state.</li>
<li>Publish your sign-in metadata.</li>
<li>Label every tool truthfully.</li>
<li>Write descriptions for humans.</li>
<li>Treat tool definitions as code.</li>
</ol>
<p>The full survey, with method, limits and every figure above, is here: <a href="https://mcpprobe.tech/research/state-of-mcp-security">State of MCP server security, November 2026</a>.</p>
<p><em>Disclosure: I built the scanner used for the survey. Only test servers you own or are authorised to test.</em></p>
]]></content:encoded></item></channel></rss>